Meet the Helmholtz AI local unit for Earth and enviroment @ HZG / DKRZ
‘Humans of Helmholtz AI’ #6: Helmholtz AI @ HZG / DKRZ
At Helmholtz AI we aim to push boundaries in terms of artificial intelligence and machine learning, but this still requires some human-powered efforts. Let us introduce the ‘humans’ who make Helmholtz AI possible!
This month, we spoke with the nine humans who currently compose the Helmholtz AI local unit for research field Earth and environment at Helmholtz-Zentrum Geesthacht (HZG) / German Climate Computing Center (DKRZ).
David Greenberg, young investigator group leader. David did his PhD in computational neuroscience, on time series and image processing problems. This led naturally to a model-based approach, where physical constraints, chemical reaction schemes and 3D geometry provided strong constraints for data processing. A postdoc at Technical University of Munich introduced him to simulation-based inference, where machine learning techniques can fit domain-specific simulators to data. Developing and applying these techniques for Earth science, one encounters a major increase in model complexity, dataset size and real-world importance of the applications.
David leads the ‘model-driven machine learning’ Helmholtz AI research group. ‘We're interested in identifying possible synergies between two different worlds of data analysis: interpretable, physics-driven modelling and data-driven machine learning. On the one hand, we want to extend the benefits of the classical physics-based approach to improve ML. At the same time, we want to use ML's capacity for efficient parameter tuning and assimilation of large and heterogeneous datasets to improve workflows involving physical models’.
Tobias Weigel, the leader of the Helmholtz AI consultant team, which is hosted by the German Climate Computing Center (DKRZ). Tobias has worked at DKRZ since 2010, contributing to and interconnecting multiple projects (mostly EU Horizon 2020 funded) in the areas of e-Science infrastructures, data management and data analysis. As part of these activities, he has been involved for many years with the Research Data Alliance (RDA), creating collaborative community standards for identifier and metadata management, and became an editorial board member of the re-launched CODATA Data Science Journal.
At Helmholtz AI, he is in charge of coordination and technical tasks, including coordination with the scientific users concerning challenges the consultants can provide expertise for, contributing to the subsequent technical work with data processing and Python coding, and fostering communication and community outreach. He states that ‘Helmholtz AI presents for me opportunities to become directly involved in a new innovative field in Data Science, and I look forward to fostering practical innovation in these areas for the benefit of the Earth science community in Helmholtz and beyond’.
Marcel Nonnenmacher, a postdoc in the model-driven machine learning group. Marcel studied cognitive science, followed by a Master's and PhD in computational neuroscience. At Helmholtz AI, he works on data-driven models for weather and climate predictions, trying to best make use of machine learning's advantages such as data integration and model differentiability.
Tobias Machnitzki, a PhD student in the model-driven machine learning group. Tobias studied Meteorology at Hamburg University, where he worked with neural networks for his Master's thesis. After graduating, he worked as a data science consultant for half a year, before coming back to the science community. At Helmholtz AI, he : ‘works on developing neural networks for environmental tasks, mainly atmospheric science’.
Caroline Arnold, a Helmholtz AI consultant at DKRZ. She holds a PhD in theoretical physics from DESY and the University of Hamburg with a thesis on ultrafast molecular dynamics. During her time at DESY, she developed scientific software, collaborated in experiments, and communicated scientific results to various audiences. At Helmholtz AI, she supports users from the Earth and climate sciences with their machine learning related projects. ‘This involves consultation on appropriate methods and models, help with data processing, and ML software engineering’ she explains.
Felix Stiehler, a Helmholtz AI consultant at DKRZ. He got a master’s degree in computer science at the Humboldt university in Berlin and then moved to another department in the university as a research scientist at the Heinrich Heine University in Düsseldorf, developing AI models for fundamental biology research. At Helmholtz AI, he works together with scientists from the Earth and climate sciences and tries to support them with all things related to AI, from data analysis and preprocessing, to model design, model training on a GPU cluster and result analysis. ‘Most of the time, I am doing applied research, but some time is also spent communicating results and catching up on the latest developments’, he says.
Jakob Lüttgau, a Helmholtz AI consultant at DKRZ. Not wanting to decide straight away in which direction the journey should go, Jakob studied computer science at the University of Hamburg and found great joy in high-performance and scientific computing. He then became a specialist for HPC middleware, I/O and storage systems, researching how to automate and optimize the interplay of available technologies and scientific workflows. ‘Machine learning workloads are challenging for many existing software and hardware solutions because they seemingly contradict optimizations that work so well for many established applications’, he explains, and so as a Helmholtz AI consultant he focuses on ‘supporting researchers from the Earth sciences to take advantage of machine learning that scales to supercomputers’.
Frauke Albrecht, a Helmholtz AI consultant at DKRZ. Frauke finished her studies of mathematics at the University of Leipzig in 2006, and through her PhD at HZG she started working in climate science, focusing on sea level change in the North Sea, a topic she further developed through two postdoc positions in Concepción, Chile, this time focusing on the South Pacific. Before joining Helmholtz AI, she participated in a Data Science Bootcamp for 3 months in order to ‘broaden my skills in data science and especially in AI’.
Pavan Kumar Siligam, a Helmholtz AI consultant at DKRZ. He holds a Master of Technology degree in Spatial Information Technology and another Masters degree in Integrated Climate System Sciences from the University of Hamburg with a thesis on “An algorithm to detect leads in sea ice”. His final Master’s semester unravelled a whole new world of ways to solve various problems using Python programming skills he had learnt during his Course work and eventually led him to be self-taught and to hone his skills. Pavan previously worked at DKRZ as a research associate; during that time, Pavan introduced the versatility of the Jupyter notebook and it’s usage and went on to teach in Workshops on Python in the Scientific Context. At Helmholtz AI, he is currently involved in investigating various scientific data-formats to benchmark the I/O throughput and to provide subsequent optimizations for the same.
Breaking research silos in an innovative, pragmatic way
According to Tobias Weigel, ‘Helmholtz AI recognises both front-line research and expert services and tries to combine these to foster innovation not just by providing funds for research projects, but also by dedicatedly supporting the role technical expertise and services have in the ML and AI field’.
He thinks this is not only an ‘innovative and pragmatic way to sharpen the focus of researchers and innovative IT experts at what they do best, but which also strongly integrates these two communities with surrounding structures that Helmholtz AI builds. Due to the complexity of the technical methods and the fast innovation also from the IT industry, I think that Helmholtz AI has a unique and very fitting setup to make best use of the new methods’.
Felix also highlights the importance of the setup and how that allows the team to ‘work on great and challenging projects related to climate science’. David is ‘really impressed with the collaborative and supportive atmosphere, both at the Institute for Coastal Research here at HZG and in the connections across centers fostered by Helmholtz AI’. As Jakob puts it, ‘Helmholtz AI breaks the silos and fosters exchange’. Tobias Machnitzki likes ‘being connected to other AI researchers, so there is always an expert at hand when I need some guidance’. Frauke agrees with him: ‘People complement each other with their knowledge and that is helpful’.
As for Caroline, she ‘particularly enjoys being involved in a variety of projects. It is nice to get a glimpse of the research endeavours across the Helmholtz Association’.
A new, lasting model to adopt AI / ML methods in research
When asked about the future of this platform, Caroline remarks that ‘many research groups are interested in pursuing ML projects, but lack the resources to actually follow through. I would envision Helmholtz AI not only filling that gap, but actively contributing by providing training and consultation to scientists’. Jakob imagines Helmholtz AI becoming ‘a model for how to handle adoption and innovation of new technologies’.
David thinks that ‘there is a great opportunity to facilitate fruitful collaborations, both in terms of common methodological interests and shared application domains’. In that sense, Tobias Weigel hopes for creating ‘a lasting impact on adoption of data science and AI methods in multiple thematic areas, illustrating clearly what can already be done with today's methods and where additional research is required. For Earth System research, I hope we can achieve a breakthrough in leveraging ML methods in understanding still difficult modelling processes such as cloud formation, atmospheric chemistry or data assimilation’.
Frauke thinks that the era of AI has already started and that ‘we can help climate science to advance in this direction. I hope that we can teach climate scientists where and how to use AI in their work and bring climate scientists and computer scientists closer together’. To that, Tobias Machnitzki adds that he sees Helmholtz AI ‘becoming an even larger initiative because of the growing field of AI’.
Felix's personal philosophy is to ‘always be ready to change your mind based on new evidence’, something that complements Tobias Weigel’s main inspiration: ‘science fiction that becomes reality’.
On more hands-on matters, Tobias Machnitzki plays guitar, both Jakob and Caroline play the piano, and Caroline sings in a choir. ‘Also, I volunteer for an organization that supports women in STEM’ she adds. Jakob mentions an interest in ‘tinkering on all sorts of little side projects, which although often unfinished, typically turn out to become valuable experiences when I least expect it’. That is probably why he says he is truly happy when he ‘finds the time to let things flow guided only by my curiosity. It is a restless beast in its own right, however!’.
Nature is one of the three things Tobias Weigel could not live without, along with ‘love and truly dark chocolate’. Felix is also a nature lover, and in that sense Caroline and Frauke like to ‘enjoy the long summer evenings while sitting at the Elbe river with a view of the busy harbour’. Dancing is another of Frauke’s passions, and there are three things Marcel cannot live without (in this order): ‘caffeine, oxygen and food.’